Carbon emission intelligent scheduling method based on multi-objective collaborative optimization

Abstract

The invention relates to a carbon emission intelligent scheduling method based on multi-objective collaborative optimization, and the method comprises the following steps: S1, recognizing and classifying different types of generator sets in a novel power system, and determining the carbon emission factor of each type of generator set according to the power generation type and efficiency through the actual measurement data and historical data; s2, establishing a dynamic carbon emission model of the novel power system; s3, obtaining predicted carbon emission based on the carbon emission prediction model, and obtaining actual carbon emission based on the dynamic carbon emission model of the novel power system; s4, constructing a carbon emission curve, and detecting whether the data is abnormal or not based on an anomaly detection model; s5, taking carbon emission as a constraint condition and an optimization target of energy distribution and scheduling, taking the carbon emission level of each unit into scheduling consideration, and constructing a multi-target scheduling model; and S6, solving the multi-target scheduling model based on a multi-target optimization algorithm Pareto, and adjusting the power generation combination to minimize the total emission of the system. The carbon emission is effectively reduced, and the energy utilization efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of electric power dispatching, and in particular to a carbon emission intelligent dispatching method based on multi-objective collaborative optimization. Background Art

[0002] In the development of modern power systems, with the increasing concern about climate change and environmental sustainability, the management of carbon emissions has become one of the focuses of the power industry. Traditional power systems mainly rely on fossil fuels for power generation, which leads to a large amount of carbon dioxide emissions. Therefore, building a new power system with low carbon as the goal has become an inevitable trend.

[0003] In recent years, new energy technologies, such as wind and solar power, have significantly increased the proportion of renewable energy generation in the power system. However, the intermittent and volatile characteristics of new energy have brought great challenges to the stable operation of the power system. In order to balance the contradiction between new energy access and system stability, more complex and intelligent scheduling and management methods need to be introduced. Summary of the invention

[0004] In order to solve the above problems, the purpose of the present invention is to provide a carbon emission intelligent scheduling method based on multi-objective collaborative optimization, which can effectively reduce carbon emissions, improve energy utilization efficiency, and enhance the flexibility and resource utilization efficiency of the power system.

[0005] To achieve the above object, the present invention adopts the following technical solutions:

[0006] A carbon emission intelligent scheduling method based on multi-objective collaborative optimization includes the following steps:

[0007] S1: Identify and classify different types of generators in the new power system, such as thermal power, wind power, photovoltaic power, etc., and determine the carbon emission factor of each type of generator based on the type and efficiency of power generation using measured data and historical data;

[0008] S2: Based on the carbon emission factors of various types of power generation units, a dynamic carbon emission model of the new power system is established;

[0009] S3: Obtaining operation data and environmental data of the new power system, and obtaining predicted carbon emissions based on a carbon emission prediction model, and obtaining actual carbon emissions based on a dynamic carbon emission model of the new power system;

[0010] S4: Combine predicted carbon emissions and actual carbon emissions to construct a carbon emission curve, and detect whether the data is abnormal based on anomaly detection models;

[0011] S5: Take carbon emissions as a constraint and optimization target for energy allocation and scheduling, incorporate the carbon emission level of each unit into scheduling considerations, and build a multi-objective scheduling model;

[0012] S6: Based on the multi-objective optimization algorithm Pareto frontier, the multi-objective scheduling model is solved to adjust the power generation mix to minimize the total system emissions while ensuring grid reliability.

[0013] Furthermore, S1 is specifically:

[0014] Obtaining operating data and technical parameters of each generator set, wherein the operating data includes power generation, fuel consumption, load level and environmental parameters, and the technical parameters include power generation technology, fuel type, equipment efficiency, emission control technology, and unit type;

[0015] Classify the generating units by type into fossil fuel units and renewable energy units, and identify the key parameters that affect carbon emissions, including fuel type, fuel carbon content, unit efficiency, and technical characteristics;

[0016] The carbon emission factor of the fossil fuel unit is calculated based on fuel consumption and power generation efficiency:

[0017]

[0018] Among them, EF fossil is the carbon emission factor of the fossil fuel unit; C is the carbon content of the fuel; H is the calorific value of the fuel; η is the thermal efficiency of the generator set, and E is the total power generation per unit time;

[0019] The indirect carbon emission factor of the renewable energy unit is estimated using the life cycle analysis method:

[0020]

[0021] Among them, EF renewable is the carbon emission factor of the renewable energy unit, LCA total is the total carbon emissions during the entire life cycle, and L is the total electricity generated during the life cycle.

[0022] Furthermore, S2 is specifically:

[0023] The carbon emission factors of fossil fuel units are modified based on different factors, as follows:

[0024] Based on the changes in fuel quality and load level, the carbon emission factors of fossil fuel units are corrected:

[0025]

[0026] Among them, EF fossil,i is the carbon emission factor of the i-th type fossil fuel unit; f load (L(t)) is the adjustment function of the load level L(t); β fuelis the fuel quality change influence coefficient; Q(t) is the fuel quality index; Q avg is the average fuel mass;

[0027] The correction of carbon emission factors of renewable energy units based on equipment performance is as follows:

[0028] EF renewable,j (t) = EF renewable,j ×f control (E(t));

[0029] Among them, f control (E(t)) is the adjustment function of equipment efficiency E(t); EF renewable,j is the carbon emission factor of the j-th type of renewable energy unit;

[0030] Construct a dynamic carbon emission model for a new power system, specifically:

[0031]

[0032] Among them, P f,i (t) is the power generation of the i-th fossil fuel unit at time t, EF fossil,i (t) is the carbon emission factor of the i-th fossil fuel unit at time t; P r,j (t) is the power generation of the j-th renewable energy unit at time t; EF renewable,j (t) is the carbon emission factor of the j-th renewable energy unit at time t.

[0033] Furthermore, the carbon emission prediction model is built based on Transformer, as follows:

[0034] Obtain the time series data of historical carbon emissions, power generation, load demand and weather data of the new power system, normalize them, and construct a time series data matrix Where T is the length of the time series, d input is the dimension of the input feature, and R represents a set of real numbers;

[0035] Convert the data input of each time step into a vector, add position information to each time step to maintain the sequence order correlation, and use the following formula for position encoding:

[0036]

[0037] Among them, pos is the position, i′ is the dimension index, and d model is the hidden dimension of the Transformer model; PE (pos,2i′) Represents the value of the 2i′th dimension at position pos; PE (pos,2i′+1) Represents the value of the 2i′+1th dimension at position pos;

[0038] The encoder layer calculates the similarity between each time step and all time steps in the sequence based on the multi-head self-attention mechanism, filters out important information in the input features, and captures different patterns of the input data through different feature combinations;

[0039] Combine the output and input of multi-head attention, stabilize the calculation through residual connection and layer normalization;

[0040] Finally, the output after the encoder is mapped to the target carbon emissions through a linear layer.

[0041] Furthermore, the multi-head self-attention mechanism is as follows:

[0042] Using a learnable parameter matrix W Q , W K , W V To generate the query Q, key K and value V:

[0043] Q=XW Q ,K=XW K ,V=XW V ;

[0044] The similarity of each pair of query and key is calculated to obtain the attention weights, and then the weighted sum is calculated using these weights:

[0045]

[0046] Among them, d k is the dimension of the key vector, the superscript T is the transpose; softmax is the normalization function, Attention(Q,K,V) is the output of the attention head;

[0047] Connect the outputs of multiple attention heads and transform them linearly:

[0048] MultiHead(Q,K,V)=Concat(head 1 ,...,head u ,...,head U )W O ;

[0049] Among them, head u is the output of the u-th attention head; W O is the output transformation matrix; MultiHead(Q,K,V) is the multi-head attention output; Concat means connection.

[0050] Furthermore, the output and input of multi-head attention are combined, and the calculation is stabilized through residual connection and layer normalization, as follows:

[0051] Combining the output and input of multi-head attention, the calculation is stabilized through residual connection and layer normalization:

[0052] Z 1 =LayNorm(X+MultiHead(Q,K,V));

[0053] Among them, LayerNorm represents layer normalization, Z 1 represents the normalized output;

[0054] Apply the same fully connected feed-forward neural network at each position:

[0055] FFN(Z 1 )=ReLU(Z 1 W 1 +b 1 )W 2 +b 2 ;

[0056] Among them, FFN is a feedforward neural network, which contains two linear transformations and one ReLU activation; W 1 ,W 2 is the weight of FFN, b 1 ,b 2 is the bias of FFN;

[0057] Using layer normalization and residual connections:

[0058] Z 2 =LayerNrom(Z1+FFN(Z1));

[0059] Among them, Z 2 is the encoder layer output;

[0060] Map the final output of the encoder layer to the target value, i.e., carbon emission prediction

[0061]

[0062] Among them, W output and b output are the parameters of the linear layer.

[0063] Furthermore, S4 is specifically:

[0064] Based on the carbon emission prediction model, the predicted carbon emissions are obtained, and based on the dynamic carbon emission model of the new power system, the actual carbon emissions are obtained;

[0065] Calculate the residual sequence Residual(t) for the collected actual and predicted carbon emission data:

[0066]

[0067] in, is the predicted value at time t, and CE(t) is the actual carbon emission data;

[0068] And based on the residual sequence and local anomaly factors, abnormal data points are detected;

[0069] Use the Matplotlib library for graphical display, draw the time series curve of carbon emissions, including forecasts and actual values, mark the abnormal points on the graph, and use different colors or shapes to emphasize them.

[0070] Furthermore, based on the residual sequence and the local anomaly factor, abnormal data points are detected as follows:

[0071] Select the number of neighbors k and calculate the k-distance, that is, calculate the distance between a sample and its kth nearest neighbor sample in the residual sequence data set. For any sample p and its neighbor o, calculate the reachability distance reach-dist k (p, o):

[0072] rech-dist k (p, o) = max (k-distance (o), distance (p, o));

[0073] Among them, k-distance(o) is the distance from o to its kth nearest neighbor, and distance(p,o) is the Euclidean distance from point p to o;

[0074] Calculate the local reachability density,

[0075]

[0076] Among them, lrd k (p) represents the neighborhood density of sample p, N k (p) represents the k nearest neighbor set of data point p;

[0077] Based on the local reachable density, calculate the local outlier factor LOF value LOF k (p):

[0078]

[0079] Among them, lrd k (o) represents the neighborhood density of sample o;

[0080] According to the LOF value, points above the set threshold are identified as outliers.

[0081] Furthermore, a multi-objective scheduling model is constructed, as follows:

[0082] The goal is to minimize the total cost of electricity generation and minimize the total carbon emissions:

[0083]

[0084] Among them, C is the total power generation cost, a i″′ , b i″′ 、c i″′ is the cost coefficient of the i′′th generator set, P i″′ is the power generation of the i″′th generator set; E is the total carbon emissions, d i″′ 、e i″′ 、f i″′ is the carbon emission coefficient of the i′′th generator set, and N is the total number of generator sets;

[0085] The constraints include power balance constraints and unit power output limits:

[0086]

[0087] Where D is the power demand of the system; P i″′ max , P i″′ min are the upper and lower limits of the power generation of the i″′th generator set respectively.

[0088] Furthermore, S6 is specifically:

[0089] S6-1: Generate an initial population, each individual represents a set of power generation combinations;

[0090] S6-2: Sort the population according to Pareto dominance and generate multiple non-dominated hierarchies;

[0091] S6-3: Calculate the crowding distance of each individual to maintain the diversity of the population;

[0092] S6-4: Use tournament selection to prefer individuals with low non-dominated levels and high crowding;

[0093] S6-5: Generation of offspring using simulated binary crossover (SBX);

[0094] S6-6: Use polynomial mutation to enhance the diversity of offspring;

[0095] S6-7: merge the parent and offspring populations and re-perform non-dominated sorting;

[0096] S6-8: Select individuals based on non-dominated sorting and crowding distance to form the next generation population;

[0097] S6-9: When the maximum number of generations or other stopping criteria are reached, the iteration is stopped and the final Pareto optimal solution set, that is, the optimal power generation combination, is returned.

[0098] The present invention has the following beneficial effects:

[0099] 1. The present invention incorporates carbon emission factors into the dispatching model, combines environmental indicators with traditional economic indicators, achieves the goal of reducing carbon emissions and improving energy efficiency, and enhances the flexibility and resource utilization efficiency of the power system;

[0100] 2. The present invention builds a carbon emission model based on Transformer, which can more accurately capture the complex time series correlation and heterogeneity characteristics in the system, thereby improving the prediction effect of carbon emissions, while maintaining a certain generalization ability and maintaining efficient calculation speed;

[0101] 3. The present invention takes carbon emissions as a constraint and optimization target for energy allocation and scheduling, incorporates the carbon emission level of each unit into scheduling considerations, and constructs a multi-objective scheduling model; adjusts the power generation mix based on the Pareto frontier of the multi-objective optimization algorithm to minimize the total system emissions while ensuring grid reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0102] Figure 1 is a flow chart of the method of the present invention;

[0103] Figure 2 Schematic diagram of the Pareto process of the multi-objective optimization algorithm in one embodiment of the present invention. DETAILED DESCRIPTION

[0104] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments:

[0105] refer to Figure 1 In this embodiment, a carbon emission intelligent scheduling method based on multi-objective collaborative optimization is provided, which is characterized by comprising the following steps:

[0106] S1: Identify and classify different types of generators in the new power system, such as thermal power, wind power, photovoltaic power, etc., and determine the carbon emission factor of each type of generator based on the type and efficiency of power generation using measured data and historical data;

[0107] S2: Based on the carbon emission factors of various types of power generation units, a dynamic carbon emission model of the new power system is established;

[0108] S3: Obtaining operation data and environmental data of the new power system, and obtaining predicted carbon emissions based on a carbon emission prediction model, and obtaining actual carbon emissions based on a dynamic carbon emission model of the new power system;

[0109] S4: Combine predicted carbon emissions and actual carbon emissions to construct a carbon emission curve, and detect whether the data is abnormal based on anomaly detection models;

[0110] S5: Take carbon emissions as a constraint and optimization target for energy allocation and scheduling, incorporate the carbon emission level of each unit into scheduling considerations, and build a multi-objective scheduling model;

[0111] S6: Based on the multi-objective optimization algorithm Pareto, the multi-objective scheduling model is solved to adjust the power generation mix to minimize the total system emissions while ensuring grid reliability.

[0112] In this embodiment, S1 is specifically:

[0113] Obtaining operating data and technical parameters of each generator set, wherein the operating data includes power generation, fuel consumption (for fossil fuel power generation), load level and environmental parameters, and the technical parameters include power generation technology, fuel type, equipment efficiency, emission control technology, and unit type;

[0114] Classify the generator sets into fossil fuel units and renewable energy units, such as coal-fired power and gas-fired power units, and wind power and photovoltaic power units as renewable energy units, and identify the key parameters that affect carbon emissions, including fuel type, fuel carbon content, unit efficiency, and technical characteristics;

[0115] The carbon emission factor of the fossil fuel unit is calculated based on fuel consumption and power generation efficiency:

[0116]

[0117] Among them, EF fossil is the carbon emission factor of the fossil fuel unit; C is the carbon content of the fuel; H is the calorific value of the fuel; η is the thermal efficiency of the generator set, and E is the total power generation per unit time;

[0118] The indirect carbon emission factors of renewable energy units (such as wind power and photovoltaic power) are estimated using the life cycle analysis method:

[0119]

[0120] Among them, EF renewable is the carbon emission factor of the renewable energy unit, LCA total is the total carbon emissions during the entire life cycle, and L is the total electricity generated during the life cycle.

[0121] In this embodiment, S2 is specifically:

[0122] The carbon emission factors of fossil fuel units are modified based on different factors, as follows:

[0123] Based on the changes in fuel quality and load level, the carbon emission factors of fossil fuel units are corrected:

[0124]

[0125] Among them, EF fossil,i is the carbon emission factor of the i-th type fossil fuel unit; f load (L(t)) is the adjustment function of the load level L(t); β fuel is the fuel quality change influence coefficient; Q(t) is the fuel quality index; Q avg is the average fuel mass;

[0126] The correction of carbon emission factors of renewable energy units based on equipment performance is as follows:

[0127] EF renewable,j (t) = EF renewable,j ×f control (E(t))

[0128] Among them, f control (E(t)) is the adjustment function of equipment efficiency E(t); EF renewable,j is the carbon emission factor of the j-th type of renewable energy unit;

[0129] Construct a dynamic carbon emission model for a new power system, specifically:

[0130]

[0131] Among them, P f,i (t) is the power generation of the i-th fossil fuel unit at time t, EF fossil,i (t) is the carbon emission factor of the i-th fossil fuel unit at time t; P r,j (t) is the power generation of the j-th renewable energy unit at time t; EF renewable,j (t) is the carbon emission factor of the j-th renewable energy unit at time t.

[0132] In this embodiment, the carbon emission prediction model is constructed based on Transformer, as follows:

[0133] Obtain the time series data of historical carbon emissions, power generation, load demand and weather data of the new power system, normalize them, and construct a time series data matrix Where T is the length of the time series, d input is the dimension of the input feature, and R represents a set of real numbers;

[0134] Convert the data input of each time step into a vector, add position information to each time step to maintain the sequence order correlation, and use the following formula for position encoding:

[0135]

[0136] Among them, pos is the position, i′ is the dimension index, and d model is the hidden dimension of the Transformer model; PE (pos,2i′) Represents the value of the 2i′th dimension at position pos; PE (pos,2i′+1) Represents the value of the 2i′+1th dimension at position pos;

[0137] The encoder layer calculates the similarity between each time step and all time steps in the sequence based on the multi-head self-attention mechanism, filters out important information in the input features, and captures different patterns of the input data through different feature combinations;

[0138] Combine the output and input of multi-head attention, stabilize the calculation through residual connection and layer normalization;

[0139] Finally, the output after the encoder is mapped to the target carbon emissions through a linear layer.

[0140] In this embodiment, the multi-head self-attention mechanism is as follows:

[0141] Using a learnable parameter matrix W Q , W K , W V To generate the query Q, key K and value V:

[0142] Q=XW Q ,K=XW K ,V=XW V ;

[0143] The similarity of each pair of query and key is calculated to obtain the attention weights, and then the weighted sum is calculated using these weights:

[0144]

[0145] Among them, d k is the dimension of the key vector, the superscript T is the transpose; softmax is the normalization function, Attention(Q,K,V) is the output of the attention head;

[0146] Connect the outputs of multiple attention heads and transform them linearly:

[0147] MultiHead(Q,K,V)=Concat(head 1 ,...,head u ,...,head U )W O ;

[0148] Among them, head u is the output of the u-th attention head; W O is the output transformation matrix;

[0149] MultiHead(Q,K,V) is the multi-head attention output; Concat means connection.

[0150] In this embodiment, the output and input of multi-head attention are combined, and the calculation is stabilized through residual connection and layer normalization, as follows:

[0151] Combining the output and input of multi-head attention, the calculation is stabilized through residual connection and layer normalization:

[0152] Z 1 =LayerNorm(X+MultiHead(Q,K,V));

[0153] Among them, LayerNorm represents layer normalization, Z 1 represents the normalized output;

[0154] Apply the same fully connected feed-forward neural network at each position:

[0155] FFN(Z 1 )=ReLU(Z 1 W 1 +b 1 )W 2 +b 2 ;

[0156] Among them, FFN is a feedforward neural network, which contains two linear transformations and one ReLU activation; W 1 ,W 2 is the weight of FFN, b 1 ,b 2 is the bias of FFN;

[0157] Using layer normalization and residual connections:

[0158] Z 2 =LayerNorm(Z1+FFN(Z1));

[0159] Among them, Z 2 is the encoder layer output;

[0160] Map the final output of the encoder layer to the target value, i.e., carbon emission prediction

[0161]

[0162] Among them, W output and boutput are the parameters of the linear layer.

[0163] In this embodiment, S4 is specifically:

[0164] Based on the carbon emission prediction model, the predicted carbon emissions are obtained, and based on the dynamic carbon emission model of the new power system, the actual carbon emissions are obtained;

[0165] Calculate the residual sequence Residual(t) for the collected actual and predicted carbon emission data:

[0166]

[0167] in, is the predicted value at time t, and CE(t) is the actual carbon emission data;

[0168] And based on the residual sequence and local anomaly factors, abnormal data points are detected;

[0169] Use the Matplotlib library for graphical display, draw the time series curve of carbon emissions, including forecasts and actual values, mark the abnormal points on the graph, and use different colors or shapes to emphasize them.

[0170] In this embodiment, based on the residual sequence and the local abnormal factor, abnormal data points are detected as follows:

[0171] Select the number of neighbors k and calculate the k-distance, that is, calculate the distance between a sample and its kth nearest neighbor sample in the residual sequence data set. For any sample p and its neighbor o, calculate the reachability distance reach-dist k (p, o):

[0172] reach-dist k (p, o) = max (k-distance (o), distance (p, o));

[0173] Among them, k-distance(o) is the distance from o to its kth nearest neighbor, and distance(p,o) is the Euclidean distance from point p to o;

[0174] Calculate the local reachability density,

[0175]

[0176] Among them, lrd k (p) represents the neighborhood density of sample p, and Nk(p) represents the k nearest neighbor set of data point p;

[0177] Based on the local reachable density, calculate the local outlier factor LOF value LOF k (p):

[0178]

[0179] Among them, lrd k (o) represents the neighborhood density of sample o;

[0180] According to the LOF value, points above the set threshold are identified as outliers.

[0181] In this embodiment, a multi-objective scheduling model is constructed, as follows:

[0182] The goal is to minimize the total cost of electricity generation and minimize the total carbon emissions:

[0183]

[0184] Among them, C is the total power generation cost, a i″′ , b i″′ 、c i″′ is the cost coefficient of the i′′th generator set, P i″′ is the power generation of the i″′th generator set; E is the total carbon emissions, d i″′ 、e i″′ 、f i″′ is the carbon emission coefficient of the i′′th generator set, and N is the total number of generator sets;

[0185] The constraints include power balance constraints and unit power output limits:

[0186]

[0187] Where D is the power demand of the system; P i″′ max , P i″′ min are the upper and lower limits of the power generation of the i″′th generator set respectively.

[0188] refer to Figure 2 In this embodiment, S6 is specifically:

[0189] S6-1: Generate an initial population, each individual represents a set of power generation combinations;

[0190] S6-2: Sort the population according to Pareto dominance and generate multiple non-dominated hierarchies;

[0191] S6-3: Calculate the crowding distance of each individual to maintain the diversity of the population;

[0192] S6-4: Use tournament selection to prefer individuals with low non-dominated levels and high crowding;

[0193] S6-5: Generation of offspring using simulated binary crossover (SBX);

[0194] S6-6: Use polynomial mutation to enhance the diversity of offspring;

[0195] S6-7: merge the parent and offspring populations and re-perform non-dominated sorting;

[0196] S6-8: Select individuals based on non-dominated sorting and crowding distance to form the next generation population;

[0197] S6-9: When the maximum number of generations or other stopping criteria are reached, the iteration is stopped and the final Pareto optimal solution set, that is, the optimal power generation combination, is returned.

[0198] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0199] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0200] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0201] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0202] The above is only a preferred embodiment of the present invention, and does not limit the present invention in other forms. Any technician familiar with the profession may use the above disclosed technical content to change or modify it into an equivalent embodiment with equivalent changes. However, any simple modification, equivalent change and modification made to the above embodiment according to the technical essence of the present invention without departing from the technical solution of the present invention still belongs to the protection scope of the technical solution of the present invention.

Claims

1. A carbon emission intelligent scheduling method based on multi-objective collaborative optimization, characterized in that: The following steps are involved: S1: Identify and classify different types of generators in the new power system, and determine the carbon emission factor for each type of generator based on the type and efficiency of power generation using measured data and historical data; S2: Based on the carbon emission factors of various types of power generation units, a dynamic carbon emission model of the new power system is established; S3: Obtaining operation data and environmental data of the new power system, and obtaining predicted carbon emissions based on a carbon emission prediction model, and obtaining actual carbon emissions based on a dynamic carbon emission model of the new power system; S4: Combine predicted carbon emissions and actual carbon emissions to construct a carbon emission curve, and detect whether the data is abnormal based on anomaly detection models; S5: Take carbon emissions as a constraint and optimization target for energy allocation and scheduling, incorporate the carbon emission level of each unit into scheduling considerations, and build a multi-objective scheduling model; S6: Based on the multi-objective optimization algorithm Pareto, the multi-objective scheduling model is solved and the power generation mix is ​​adjusted to minimize the total system emissions.

2. According to claim 1, a carbon emission intelligent scheduling method based on multi-objective collaborative optimization is characterized in that: The S1 is specifically: Obtaining operating data and technical parameters of each generator set, wherein the operating data includes power generation, fuel consumption, load level and environmental parameters, and the technical parameters include power generation technology, fuel type, equipment efficiency, emission control technology, and unit type; Classify the generator sets into fossil fuel units and renewable energy units according to their types, and identify the key parameters that affect carbon emissions, including fuel type, fuel carbon content, unit efficiency, and technical characteristics; the carbon emission factors of fossil fuel units are calculated based on fuel consumption and power generation efficiency: Among them, EF fossil is the carbon emission factor of the fossil fuel unit; C is the carbon content of the fuel; H is the calorific value of the fuel; η is the thermal efficiency of the generator set, and E is the total power generation per unit time; The indirect carbon emission factor of the renewable energy unit is estimated using the life cycle analysis method: Among them, EF renewable is the carbon emission factor of the renewable energy unit, LCA total is the total carbon emissions during the entire life cycle, and L is the total electricity generated during the life cycle.

3. According to claim 2, a carbon emission intelligent scheduling method based on multi-objective collaborative optimization is characterized in that: The S2 is specifically: The carbon emission factors of fossil fuel units are modified based on different factors, as follows: Based on the changes in fuel quality and load level, the carbon emission factors of fossil fuel units are corrected: Among them, EF fossil,i is the carbon emission factor of the i-th type fossil fuel unit; f load (L(t)) is the adjustment function of the load level L(t); β fuel is the fuel quality change influence coefficient; Q(t) is the fuel quality index; Q avg is the average fuel mass; The correction of carbon emission factors of renewable energy units based on equipment performance is as follows: EF renewable,j (t)=EF renewable,j ×f control (E(t)) Among them, f control (E(t)) is the adjustment function of equipment efficiency E(t); EF renewable,j is the carbon emission factor of the j-th type of renewable energy unit; Construct a dynamic carbon emission model for a new power system, specifically: Among them, P f,i (t) is the power generation of the i-th fossil fuel unit at time t, EF fossil,i (t) is the carbon emission factor of the i-th fossil fuel unit at time t; P r,j (t) is the power generation of the j-th renewable energy unit at time t; EF renewable,j (t) is the carbon emission factor of the j-th renewable energy unit at time t.

4. According to claim 1, a carbon emission intelligent scheduling method based on multi-objective collaborative optimization is characterized in that: The carbon emission prediction model is built based on Transformer, as follows: Obtain the time series data of historical carbon emissions, power generation, load demand and weather data of the new power system, normalize them, and construct a time series data matrix Where T is the length of the time series, d input is the dimension of the input feature, and R represents a set of real numbers; Convert the data input of each time step into a vector, add position information to each time step to maintain the sequence order correlation, and use the following formula for position encoding: Among them, pos is the position, i′ is the dimension index, and d model is the hidden dimension of the Transformer model; PE (pos,2i′) Represents the value of the 2i′th dimension at position pos; PE (pos,2i′+1) Represents the value of the 2i′+1th dimension at position pos; The encoder layer calculates the similarity between each time step and all time steps in the sequence based on the multi-head self-attention mechanism, filters out important information in the input features, and captures different patterns of the input data through different feature combinations; Combine the output and input of multi-head attention, stabilize the calculation through residual connection and layer normalization; Finally, the output after the encoder is mapped to the target carbon emissions through a linear layer.

5. The carbon emission intelligent scheduling method based on multi-objective collaborative optimization according to claim 4 is characterized in that: The multi-head self-attention mechanism is as follows: Using a learnable parameter matrix W Q , W K , W V To generate the query Q, key K and value V: Q=XW Q ,K=XW K ,V=XW V ; The similarity of each pair of query and key is calculated to obtain the attention weights, and then the weighted sum is calculated using these weights: Among them, d k is the dimension of the key vector, the superscript T is the transpose; softmax is the normalization function; Attention(Q,K,V) is the output of the attention head; Connect the outputs of multiple attention heads and transform them linearly: MultiHead(Q,K,V)=Concat(head1,...,head u ,...,head U )W O ; Among them, head u is the output of the u-th attention head; W O is the output transformation matrix; MultiHead(Q,K,V) is the multi-head attention output; Concat means connection.

6. The carbon emission intelligent scheduling method based on multi-objective collaborative optimization according to claim 5 is characterized in that: The output and input of the combined multi-head attention are stabilized by residual connection and layer normalization, as follows: Combining the output and input of multi-head attention, the calculation is stabilized through residual connection and layer normalization: Z1=LayerNorm(X+MultiHead(Q,K,V)); Among them, LayerNorm represents layer normalization, and Z1 represents the normalized output; Apply the same fully connected feed-forward neural network at each position: FFN(Z1)=ReLU(Z1W1+b1)W2+b2; Among them, FFN is a feedforward neural network, which contains two linear transformations and one ReLU activation; W1, W2 are the weights of FFN, b1, b2 are the biases of FFN; Using layer normalization and residual connections: Z2 = LayerNorm(Z1 + FFN(Z1)); Among them, Z2 is the encoder layer output; Map the final output of the encoder layer to the target value, i.e., carbon emission prediction Among them, W output and b output are the parameters of the linear layer.

7. The carbon emission intelligent scheduling method based on multi-objective collaborative optimization according to claim 1 is characterized in that: The S4 is specifically: Based on the carbon emission prediction model, the predicted carbon emissions are obtained, and based on the dynamic carbon emission model of the new power system, the actual carbon emissions are obtained; Calculate the residual sequence Residual(t) for the collected actual and predicted carbon emission data: in, is the predicted value at time t; CE(t) is the actual carbon emission data; And based on the residual sequence and local anomaly factors, abnormal data points are detected; Use the Matplotlib library for graphical display, draw the time series curve of carbon emissions, including forecasts and actual values, mark the abnormal points on the graph, and use different colors or shapes to emphasize them.

8. The carbon emission intelligent scheduling method based on multi-objective collaborative optimization according to claim 7 is characterized in that: The abnormal data points are detected based on the residual sequence and the local abnormal factor, as follows: Select the number of neighbors k and calculate the k-distance, that is, calculate the distance between a sample and its kth nearest neighbor sample in the residual sequence data set. For any sample p and its neighbor o, calculate the reachability distance reach-dist k (p,o): reach-dist k (p,o)=max(k-distance(o),distance(p,o)); Among them, k-distance(o) is the distance from o to its kth nearest neighbor, and distance(p,o) is the Euclidean distance from point p to o; Calculate the local reachability density, Among them, lrd k (p) represents the neighborhood density of sample p, N k (p) represents the k nearest neighbor set of data point p; Based on the local reachable density, calculate the local outlier factor LOF value LOF k (p): Among them, lrd k (o) represents the neighborhood density of sample o; According to the LOF value, points above the set threshold are identified as outliers.

9. The carbon emission intelligent scheduling method based on multi-objective collaborative optimization according to claim 1 is characterized in that: The multi-objective scheduling model is constructed as follows: The goal is to minimize the total cost of electricity generation and minimize the total carbon emissions: Among them, C is the total power generation cost, a i″′ , b i″′ 、c i″′ is the cost coefficient of the i′′th generator set, P i″′ is the power generation of the i″′th generator set; E is the total carbon emissions, d i″′ 、e i″′ 、f i″′ is the carbon emission coefficient of the i′′th generator set, and N is the total number of generator sets; The constraints include power balance constraints and unit power output limits: Where D is the power demand of the system; are the upper and lower limits of the power generation of the i″′th generator set respectively.

10. The carbon emission intelligent scheduling method based on multi-objective collaborative optimization according to claim 9 is characterized in that: The S6 is specifically: S6-1: Generate an initial population, each individual represents a set of power generation combinations; S6-2: Sort the population according to Pareto dominance and generate multiple non-dominated hierarchies; S6-3: Calculate the crowding distance of each individual to maintain the diversity of the population; S6-4: Use tournament selection to prefer individuals with low non-dominated levels and high crowding; S6-5: Generation of offspring using simulated binary crossover; S6-6: Use polynomial mutation to enhance the diversity of offspring; S6-7: merge the parent and offspring populations and re-perform non-dominated sorting; S6-8: Select individuals based on non-dominated sorting and crowding distance to form the next generation population; S6-9: When the maximum number of generations or other stopping criteria are reached, the iteration is stopped and the final Pareto optimal solution set, that is, the optimal power generation combination, is returned.

Images